feat(qwen_vl): add Qwen3-VL-8B-Instruct

families/qwen_vl/tests/manifests/qwen3-vl-2b.json

@@ -1,6 +1,7 @@
 {
   "name": "qwen3-vl-2b",
   "hf_id": "Qwen/Qwen3-VL-2B-Instruct",
+  "hf_revision": "89644892e4d85e24eaac8bacfd4f463576704203",
   "bundle": "qwen3-vl-2b.bundle",
   "family": "qwen_vl",
   "task": "vision_language_generation",

families/qwen_vl/tests/manifests/qwen3-vl-8b.json

@@ -0,0 +1,22 @@
+{
+  "name": "qwen3-vl-8b",
+  "hf_id": "Qwen/Qwen3-VL-8B-Instruct",
+  "hf_revision": "0c351dd01ed87e9c1b53cbc748cba10e6187ff3b",
+  "bundle": "qwen3-vl-8b.bundle",
+  "family": "qwen_vl",
+  "task": "vision_language_generation",
+  "precision": "bf16",
+  "reference_precision": "bf16",
+  "testcases": [
+    {
+      "name": "qwen3-vl-8b",
+      "premerge": true,
+      "prompt": "What color is the vehicle in this image? Answer in one word.",
+      "max_new_tokens": 10,
+      "test_image": "data/test_img.jpeg",
+      "reference_precision": "bf16"
+    }
+  ],
+  "tensor_parallel_size": 1,
+  "max_sequence_length": 256
+}

qualification_tests/benchmark_qualification/performance/config/release.yaml

@@ -137,6 +137,11 @@ excluded_profiles:
       nano width, which is the only one whose class-attention norm covers a
       single token, but the release-performance workload and receipt were
       collected only for xcit_tiny_12_p16_224.
+  - model: qwen3-vl-8b
+    reason: >-
+      Functional and Hugging Face reference-parity qualification is present,
+      but the release-performance workload and receipt were collected only for
+      qwen3-vl-2b, which exercises the same builder and runtime path.
 
 entries:
   - id: detr.detect

website/data/hf-model-metadata.json

@@ -300,6 +300,17 @@
       ],
       "architecture_source": "config.architectures"
     },
+    {
+      "hf_id": "Qwen/Qwen3-VL-8B-Instruct",
+      "revision": "0c351dd01ed87e9c1b53cbc748cba10e6187ff3b",
+      "revision_source": "resolved",
+      "metadata_file": "config.json",
+      "model_type": "qwen3_vl",
+      "architectures": [
+        "Qwen3VLForConditionalGeneration"
+      ],
+      "architecture_source": "config.architectures"
+    },
     {
       "hf_id": "Qwen/Qwen3.5-9B",
       "revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",